Search Web Corpus

   - use wildcards / alternatives to match more than one word-form or phrase
   - select correct database table for the number of words to match
English only!
   in this database... Search Web for concordances of specific words or phrases

WC2007 WC2006 WaCUsers WaCMarginal WaCJunk

Web Corpus 2007, release 1: 3,123,996 types, 518,129,710 tokens
1-grams: 3,123,996 types, all occurrences
2-grams: 11,376,755 occur 3 or more times
3-grams: 21,671,269 occur 3 or more times
4-grams: 17,888,245 occur 3 or more times
5-grams: 10,299,890 occur 3 or more times
6-grams: 6,177,977 occur 3 or more times

Web Corpus 2006: 950,087 types, 104,410,599 tokens
1-grams dirty: 1,420,092 types, 139,565,882 tokens, more noise
1-grams: 950,087 types
2-grams: 16,066,461 types
3-grams: 51,094,827 types
4-grams: 79,115,008 types
5-grams: 19,665,341 types; some data missing
6-grams: 35,697,213 types

Web as Corpus Users: 1,090,414 types, 148,505,932 tokens, derived from Web as Corpus users' search results Oct 07-Feb 08 (English only)
1-grams: 1,090,414 types, all occurrences
2-grams: 2,215,897 occur 3 or more times
3-grams: 3,170,042 occur 3 or more times
4-grams: 2,325,161 occur 3 or more times
5-grams: 1,570,464 occur 3 or more times
6-grams: 1,243,662 occur 3 or more times

Web as Corpus Users Marginal: 309,749 types, 8,887,168 tokens, derived from Web as Corpus users' search results not identified unambiguously as English text by an automatic algorithm
1-grams: 309,749 types, all occurrences
2-grams: 456,572 occur 3 or more times
3-grams: 396,464 occur 3 or more times
4-grams: 241,411 occur 3 or more times
5-grams: 175,196 occur 3 or more times
6-grams: 150,745 occur 3 or more times

Web as Corpus Junk: 502,914 types, 13,297,906 tokens, derived from Web as Corpus users' search results rejected as English text by an automatic algorithm. Includes non-English texts, lists, program code, search-engine spam...
1-grams: 502,914 types, all occurrences
2-grams: 689,057 occur 3 or more times
3-grams: 632,528 occur 3 or more times
4-grams: 436,083 occur 3 or more times
5-grams: 343,214 occur 3 or more times
6-grams: 302,969 occur 3 or more times

Max. types to show: 1000 | 5000 | 10,000 | 50,000

Wildcards
% matches 0 or more chars, e.g. ex% matches ex, expression, ex-president... (alternative: *)
_ matches exactly 1 char, e.g. s_ng matches sang, sing, song, sung (alternative: ?)

Alternatives
Use | for OR, e.g. %gram|%gramme matches program, programme...

Exclusion
Use ~ for NOT, e.g. %gram ~%-gram matches program, ngram... but not 3-gram, n-gram...

Filter 1-grams Show all matching forms (fastest): No filtering
Show primarily innovative forms option 1: filters out forms of lemmas occurring in the Oxford Advanced Learner's Dictionary of Current English in David Hardcastle's extension of CUVPlus.
Show primarily innovative forms option 2: filters out (bio)medical terms as well as many general wordforms in the National Institutes' of Health National Library of Medicine's SPECIALIST lexicon.
Show primarily innovative forms option 3: filters out forms of lemmas occurring in either of the above repertories of established forms.
Show primarily innovative forms option 4 (slower): filters out over half a million wordforms in the British National Corpus
Show primarily innovative forms option 5 (slowest): filters out 787,755 word-forms occurring in any of the above sources.

Several wildcard-searchable Web-based corpora are accessible from this page:

Web Corpus 2007, compiled in July 2007. The first iteration comprises 3,123,996 types and 518,129,710 tokens. The goal is to produce a corpus of at least a billion words annotated with the same Part of Speech tagset as the British National Corpus.
Web Corpus 2006, based on a corpus of about 104 M tokens 'clean' version (140 M tokens 'dirty' version) compiled from the Web in Feb-Mar 2006. (The original webpages on which these datasets are based were lost in a hard disk crash, so there are some gaps in the data.)
WaC Users, derived from search results of users of this site's Web Concordancer. Based towards users' interest, with no claim to breadth or representativeness.
WaC Users Marginal, based on text chunks in users' search results not unambiguously identified as English text.
WaC Users Junk, based on text chunks in users' search results rejected as English text. Contains fragments, lists, search-engine spam, non-English content, and occasional gems.

Some pre-built queries

Word forms 40 characters or longer (includes some digital noise)
Long forms with 3 or more hyphens
Long forms with 3 or more periods = full stops = dots

WebAsCorpus.org English Web Corpus Family find 1-6-grams

Some pre-built queries

WebAsCorpus.org English Web Corpus Family
find 1-6-grams